Radio interferometry
A Socratic walk-through of radio interferometry — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why can two small dishes placed far apart resolve finer detail than one enormous dish between them?
Two dishes, each a few tens of metres across, sitting a hundred kilometres apart. Between them, most of the sky's radio waves fall on empty ground and are lost. Yet that pair resolves finer detail than a single dish a hundred metres wide would — finer, in fact, than any dish anyone could build.
The obvious reading is that collecting more must be what buys detail, and the pair collects almost nothing by comparison. So collecting cannot be the thing that buys detail. What is?
Reasoning it through
REASONING #Begin with why big apertures resolve at all. The angular detail a telescope can separate is set, to within a factor near one, by the wavelength divided by the aperture width — and that ratio is unforgiving in the radio. Visible light has a wavelength under a micron; a radio astronomer might work at twenty-one centimetres, a couple of hundred thousand times longer. Matching even a modest optical telescope there would need an aperture measured in kilometres, and steerable dishes top out near a hundred metres because a bigger one sags under its own weight. The obvious route is closed.
So ask what a dish is doing, mechanically. A wavefront arrives across its surface, and the dish's curve delays each part of it by exactly the amount needed to bring them all into step at the focus. On axis they reinforce; slightly off axis they arrive out of step and cancel. That cancellation is the resolution — and what produces it is not the metal but the comparison between pairs of points on the surface, above all the widest-separated pairs, whose relative delay changes fastest as the source moves off axis.
That is the crack in the folk account. A dish compares every pair of points across its face at once, and its resolving power comes from its widest pair. The metal in between is doing a different job: gathering photons, which is sensitivity, not sharpness.
So why not keep the widest pair and discard the middle? That is an interferometer. Point two dishes at the same source, record what each receives, and combine the records so they interfere. Move off axis and the extra path to one dish lengthens, swinging the combination from reinforcement to cancellation. One fringe is the wavelength divided by the separation — the baseline — not by the dish size. Separation is a surveying problem rather than an engineering one, and surveying scales far better than steel.
Put numbers on it. At twenty-one centimetres over a thirty-six-kilometre baseline, wavelength divided by baseline is about six millionths of a radian — a little over one arcsecond, roughly what a large array delivers in its most extended configuration. Push further: at 1.3 millimetres across a baseline the size of Earth, some twelve thousand seven hundred kilometres, the same division gives about a ten-billionth of a radian, near twenty microarcseconds. That is the scale at which the Event Horizon Telescope resolved a black hole's shadow, and it falls out of one division.
But one baseline gives one number, not a picture. What does a single pair measure? How much structure the sky has at one particular scale and orientation — one Fourier component of the brightness distribution. A source smooth on that scale produces no fringe; a source structured on it produces a strong one. Collect many baselines of different lengths and directions, and the image is reconstructed from the set. Earth's rotation helps for free, slowly swinging every baseline's orientation as seen from the source. That is aperture synthesis, recognised with Ryle's Nobel share in 1974.
The analogy
THE ANALOGY #Think of locating a sound by ear rather than by eye. Two microphones set far apart hear it with a slight difference in arrival time, and how that difference changes as the source shifts gives the direction very precisely — more precisely, the further apart you set them. Add a third and fourth at other separations and you can tell one source from two close together, and eventually from a whole row. The precision came from the spacing between microphones; the loudness came from how big each microphone was.
Ears locate a sound but do not image it, whereas an interferometer must combine many baselines and then solve for a picture — and that solving step is a reconstruction with assumptions in it, which has no counterpart in listening.
Clarifying the model
THE MODEL #The trade is real, and glossing over it is the most common way this is oversold.
The array has the resolution of a dish as wide as its longest baseline, but the sensitivity of only the metal actually present. Faint sources need collecting area, and no amount of separation supplies it. That is why arrays keep adding elements rather than merely spreading further apart.
More subtly, an array is blind at the scales it does not sample. Since two dishes cannot occupy the same spot, no baseline is shorter than a dish diameter, so the very largest angular scales — smooth, extended emission — produce no fringes anywhere in the array and are simply absent from the image. This "missing short spacings" problem is why smooth structure often has to be filled in from a single large dish observing the same field. An interferometer image is therefore not a photograph but an inference from incomplete sampling, and the algorithms that make it are a genuine part of the instrument.
Finally, the combination must be done to a fraction of a wavelength. Nearby elements share a cable and a clock; across continents that is impossible, so each station time-stamps its stream with a hydrogen maser and the signals are correlated later. That is why very long baseline results appear months after the observation, and why the atmosphere's fluctuating delay sets the practical limit on how short a wavelength can be used.
A picture of it
THE PICTURE #How to readRead downward as time. The first two arrows are the whole mechanism: the same wavefront reaches the two dishes at different moments, and the size of that difference is what encodes direction. The self-arrows on the correlator are processing steps rather than exchanges — aligning, multiplying, and finally reconstructing. Note that step seven yields a single number about the sky, not a picture; the picture only appears at the last step, after the baseline has been swung through many orientations.
What became clearer
WHAT CLEARED #Resolution is not bought with collecting area at all — it is bought with the separation between the points being compared. A filled dish quietly does both jobs at once, which is why they get conflated: its width sets its sharpness and its area sets its sensitivity. Separate the jobs and they can be paid for separately. An interferometer buys sharpness cheaply by moving small collectors far apart, and pays in sensitivity and in the angular scales it cannot see — less a telescope than a machine for measuring the sky one spatial frequency at a time.
Where to go next
ONWARD #- How images are actually reconstructed from incomplete sampling, and what artefacts the reconstruction can invent.
- Why the same principle in optical light demands nanometre-level path control and so remains far harder than in the radio.
Key terms
TERMS #| Term | What it means |
|---|---|
| Baseline | the separation between two array elements, which sets the angular scale of the fringes they produce. |
| Fringe | the alternating reinforcement and cancellation produced when two elements' signals are combined. |
| Aperture synthesis | building up coverage of many baselines, aided by Earth's rotation, so an image can be reconstructed. |
| Very long baseline interferometry | correlating independently recorded, atomic-clock-stamped data from stations too far apart to link directly. |
Every term the collection defines is gathered in the glossary.